---
id: con-delegating-to-ai
schema: concept/1
layer: garden
status: emerging
name: Delegating to AI
aka: ["handing work to an agent", "handover brief", "is it worth delegating", "management as an AI skill"]
summary: >
  Deciding whether a task is worth handing to AI, by weighing how long it takes you against how often
  the AI gets it right and how long asking and checking take, then handing it over with a brief like
  you'd give a colleague. Framed this way by Ethan Mollick.
first_seen: "2026-01"
relations:
  builds_on: [con-jagged-frontier, con-agents]
  evolved_from: []
  related: [con-steering, con-verification, con-solution-ladder]
  parallels:
    - idea: "a creative brief"
      field: "advertising"
      note: "One page that says what the work is for, who it is for, what it must include and when it is due, written before anyone starts."
    - idea: "the five-paragraph order"
      field: "military"
      note: "Situation, mission, how it will be done, support and command: a fixed shape so nothing important is left out of a handover."
implementations: []
competencies: [cmp-describe-the-task, cmp-steering, cmp-check-the-result]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "The map had agents, steering and human in the loop but no card for the handover decision itself, the first thing people meet as they move from chat to agents."
---

In "Management as AI superpower" (January 2026), Ethan Mollick offers a quick test for whether to hand a task to AI. Compare three things: how long the task takes you, how often the AI gets this kind of task right, and how long it takes to ask, wait and check. If checking takes almost as long as doing, or the AI often fails at this kind of task, keep it.

If you do hand it over, brief it the way you'd brief a new colleague. Mollick's list: the goal and why it matters, limits on what the AI may do, what done looks like, the outputs, checkpoints, and a self-check before it hands back.

## Example

A volunteer coordinator considers asking an agent to sort 200 sign-up forms by availability. It would take her two hours; checking a sample takes fifteen minutes. She writes a five-line brief, including "flag any form with no phone number, don't guess", and hands it over.

## How it connects

- Builds on the **jagged frontier** (how often will it succeed?) and **agents** (what can it do alone?).
- **Steering** is what you do while the work runs; **verification** comes after it.
- Sometimes the answer is a simpler rung on the **solution ladder**.

## Try it

Pick one task you plan to give AI. Write down your three numbers: minutes to do it yourself, a guess at how often AI gets it right, and minutes to check. Decide, then write the brief.